Fix schema build for nested dataclasses / TypedDicts with discriminators

[sydney-runkle]

Revisit the changes made in #7646, which introduced a few bugs for edge cases when building schemas for nested dataclasses / TypedDicts with discriminators.

Fix #8271
Fix #8628

Selected Reviewer: @hramezani

[cloudflare-pages]

Deploying with    Cloudflare Pages

Latest commit:	5c151c5
Status:	✅  Deploy successful!
Preview URL:	https://2a687d8e.pydantic-docs2.pages.dev
Branch Preview URL:	https://fix-discriminators-for-datac.pydantic-docs2.pages.dev

View logs

[sydney-runkle]

Please review

[codspeed-hq]

CodSpeed Performance Report

Merging #8950 will not alter performance

_{Comparing fix-discriminators-for-dataclasses (5c151c5) with main (b09e06c)}

Summary

✅ 10 untouched benchmarks

[adriangb]

If this fixes the new test cases and doesn't break others this seems good to me. Great job on the comments by the way!

Do we have any sort of benchmark tests for this stuff? My main concern with future changes is that evidently modifying 1 line of code can take us from linear to exponential complexity. Maybe we can add a really naive test that just calls time() on 1 input and time() on 1000 input and make sure it's <= 5k times slower (so give or take linear)?

[sydney-runkle]

@adriangb,

Good question. To my knowledge, we don't have extensive performance testing for these things. I've opened an issue here: #8953. I'd like to address this issue before the patch release so that we can verify that we've improved things significantly.